U.S. and China experts propose nuclear-style safeguards for AI risks